A computing power platform operation and maintenance data analysis system and method based on artificial intelligence
By analyzing customer service response records and user interaction data, predicting the keywords and deviation coefficients of the user's question content, interrupting invalid interactions, solving the phenomenon that artificial intelligence assistants cannot accurately understand user problems, and improving the effectiveness and user experience of answers.
Patent Information
- Application Number
- CN202410661948.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-27
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2044-05-27
AI Technical Summary
In the prior art, artificial intelligence assistants cannot fully and accurately understand the content of user questions, resulting in repeated communication between users and artificial intelligence assistants on the same issue, unable to effectively express their needs, and hindering the user's help process.
By extracting historical records from the customer service response record database, obtaining keywords of the task node, predicting the content of the user's question again, calculating the deviation coefficient and characteristic word overlap of the question content, and interrupting the user and the artificial intelligence assistant when the retention coefficient is greater than the threshold. The interaction of the large language model is trained to improve the effectiveness of the answer.
It improves the accuracy of prediction of user questions, judges the effectiveness of answers from artificial intelligence assistants, reduces the occurrence of ineffective communication among users, and optimizes the interaction process between users and artificial intelligence assistants.
Smart Images

Figure CN118484515B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data management technology, and specifically to a computing power platform operation and maintenance data analysis system and method based on artificial intelligence. Background Art
[0002] The Large Language Model (LLM) is a deep learning model trained on massive text data. The large language model and its application in the field of artificial intelligence have become a hot topic in global scientific research. By inputting a specific data set and combining it with technologies such as machine learning and natural language processing, the output content and output method of the large language model can be adjusted to solve problems in specific fields. For example, training can be performed based on a small-scale data set in a specific field to customize solutions to specific problems in different fields such as retail, finance, and entertainment.
[0003] In the prior art, an artificial intelligence assistant is obtained through training of a large language model to analyze and classify the questions raised by users. On the one hand, it helps users to solve related problems by themselves, and on the other hand, it helps customer service personnel to pre-process users' questions to reduce the workload of manual customer service. In the general customer service workflow, the artificial intelligence assistant is deployed before the manual customer service, and the user will communicate with the artificial intelligence assistant first after entering the customer service system. The current artificial intelligence assistant cannot fully and accurately understand the meaning of the user's questions. There is a situation where the user cannot get the answer they want after asking the question, resulting in repeated communication between the user and the artificial intelligence assistant on the same issue. The user cannot further express his needs, which hinders the user's help process. Summary of the invention
[0004] The purpose of the present invention is to provide a computing power platform operation and maintenance data analysis system and method based on artificial intelligence to solve the problems raised in the above background technology.
[0005] In order to solve the above technical problems, the present invention provides the following technical solution: a computing power platform operation and maintenance data analysis method based on artificial intelligence.
[0006] Step S100: Obtain the historical records of processing answering tasks from the customer service answering record database, obtain the task nodes in the answering task process, extract the node features in each task node from the historical data, and obtain the keywords corresponding to each task node;
[0007] Step S200: In an interactive system that uses an artificial intelligence assistant for answering, collect the historical records of data interaction between the user and the artificial intelligence assistant, extract the pattern of the user asking questions again after the artificial intelligence assistant answers, and predict the keywords included in the content of the user's questions again;
[0008] Step S300: a user currently communicating with the artificial intelligence assistant is set as a target user, keywords included in the first predicted content are obtained according to the first question content of the target user, and a deviation coefficient between the first question content and the first predicted content is calculated for the second question content;
[0009] Step S400: The target user's multiple questions are grouped into a question sequence, and the question sequence is divided by unit sampling period. When the number of newly added feature words in the second unit sampling period is less than that in the first unit sampling period, the feature word overlap between the question contents in the second unit sampling period is calculated;
[0010] Step S500: Obtain the deviation coefficient and feature word overlap of the question content in a unit sampling period, calculate the retention coefficient, and when the retention coefficient corresponding to the second unit sampling period is greater than the management threshold, obtain the current task node of the target user, and compare the question content in the third unit sampling period with that in the second unit sampling period. When the question content includes no increase in keywords for the next task node, the interaction between the target user and the artificial intelligence assistant is interrupted at the end of the third unit sampling period.
[0011] Furthermore, step S100 includes:
[0012] Step S101: obtaining the user's needs from the customer service response record, arranging the interaction data of the same user in chronological order, obtaining an interaction sample sequence, the interaction data including text, voice or video interaction data, and collecting the interaction sample sequences of different users with the same user need to obtain an interaction sample set corresponding to the user need;
[0013] Step S102: Obtaining task nodes in the processing flow of each user demand, extracting interaction data corresponding to each task node from the interaction sample set, and extracting feature words corresponding to the task node from the interaction data, wherein the feature words are keywords corresponding to the task node;
[0014] Keywords are used as the training set of prompt words for the large speech model to train the large language model. The prompt words are used as an incentive for the large language model. When the large language model receives effective prompt words, it will have better output effects.
[0015] Further, step S200 includes:
[0016] Step S201: Obtain a first task node and a second task node from a processing flow of any task requirement, where the first task node and the second task node are two adjacent task nodes, and obtain a first keyword set composed of characteristic words corresponding to the first task node and a second keyword set composed of characteristic words corresponding to the second task node;
[0017] Step S202: collect all interaction sample sequences including the first task node and the second task node in the historical records to form a target sample sequence set, obtain any keyword prompt1 from the first keyword set, obtain the number K1 of interaction sample sequences including prompt1 in all interaction sample sequences, obtain any keyword prompt2 from the second keyword set, and when an interaction sample sequence includes prompt1 and prompt2, construct a target keyword pair: prompt1→prompt2;
[0018] Step S203: obtaining the number K2 of target keyword pairs in the target sample sequence set, and calculating the occurrence probability of keyword prompt2 α=K2 / K1;
[0019] Calculate the probability of a keyword in the second task node when a keyword corresponding to the first task node appears, obtain all the keywords of the second task node corresponding to prompt1 in the historical records, and calculate the probability of prompt2 from them;
[0020] Step S204: Obtain the target user's interaction data at the first task node, collect the keywords in the interaction data into a set H, and obtain the cumulative occurrence probability of any keyword corresponding to the second task node, where the probability cumulative value of the i-th keyword corresponding to the second task node is R i , Among them, α j represents: the probability of occurrence of any keyword of the i-th node in the second task when the j-th keyword in the set H appears, and b represents the number of keywords in H;
[0021] Analyze the interaction data of the target user in the first task node, parse out several prompt words of the first task node, extract the specific needs of the user, and make a more accurate prediction of the keywords that may appear in the second task node according to the specific needs of the user;
[0022] Step S205: Gather the probability cumulative values of the keywords corresponding to the second task nodes corresponding to the keywords in H, set a probability management threshold, and add key words with probability cumulative values greater than or equal to the probability management threshold into the keyword prediction set.
[0023] Furthermore, step S300 includes:
[0024] Step S301: obtaining any question data of the target user at the first task node as the first question content, collecting the keywords in the first question content into a set H1, and obtaining the keyword prediction set Q1 corresponding to the second task node according to H1;
[0025] Step S302: Obtain a piece of question data after the first question content as the second question data, collect the keywords in the second question content into a set H2, calculate the similarity between the keywords in H2 and the keywords in H1, record it as the first keyword similarity η1, calculate the similarity between the keywords in H2 and the keywords in Q1, record it as the second keyword similarity η2;
[0026] Step S303: Calculate the deviation coefficient y corresponding to the first question content and the second question content, y=η1-η2;
[0027] When the user asks the second question, after receiving the answer to the first question, he asks a new question, and compares the keywords in the second question with the keywords in the first question to see if the second question is more inclined to the keywords in the second task node or the keywords in the first question.
[0028] When the second question content is more inclined to the keywords in the second task node, it means that the answer received by the user is an effective answer, and the answer provided by the artificial intelligence assistant helps the user to further solve the problem. When the second question content is more inclined to the keywords in the first question content, it means that the answer received by the user does not effectively help the user solve the problem.
[0029] In order to improve the accuracy of user tendency analysis, it is necessary to perform pre-screening in step S200 on the keywords corresponding to the second task node;
[0030] y represents the degree to which the second question content deviates from the first question content. When y is larger, the effectiveness of the information received by the user is lower.
[0031] Furthermore, step S400 includes:
[0032] Step S401: Extract features from any question content and obtain feature words corresponding to each question content. In the question sequence, the feature words corresponding to the pth question content form a set E p , the feature words corresponding to the p+1th question content form a set E p+1 , calculate the feature word difference set D, D = E p+1 -E p , get the number of feature words numD in D;
[0033] Step S402: summing the number of feature words in all feature word difference sets in any unit sampling period, and calculating the average value of the summation results, where the average value is the number of newly added feature words corresponding to any unit period;
[0034] Step S403: Arrange the feature words corresponding to each question content in any unit sampling period to obtain L, L: f1, f2, f3, ... fn, where f1, f2, f3, ... and fn respectively represent the set of feature words corresponding to the 1st, 2nd, 3rd, ... and nth question content in any unit sampling period;
[0035] Step S404: Take the intersection of all sets in L to obtain set L1, take the union of all sets in L to obtain set L2, obtain the number of feature words numl1 in set L1, the number of feature words numl2 in set L2, and calculate the feature word overlap degree β in any unit sampling period, β=numL1 / numL2;
[0036] Analyze the content of questions in a certain unit sampling period. If similar content appears repeatedly, it means that the user receives fewer effective prompts.
[0037] Furthermore, step S500 includes:
[0038] Step S501: sum the deviation coefficients in the second unit sampling period to obtain the first deviation value Y, calculate the retention coefficient reten of the second unit sampling period, reten=Y×β, obtain the management threshold μ, and when reten≥μ, obtain the keywords corresponding to the second task node in all question contents in the first unit sampling period to form a set G1, and obtain the keywords corresponding to the second task node in all question contents in the second unit sampling period to form a set G2;
[0039] Step S502: Calculate the first keyword difference set Z1, Z1 = G2-G1, obtain the keywords in all the question contents in the third unit sampling period, form a set G3, and calculate the second keyword difference set Z2, Z2 = G3-G2;
[0040] Step S503: Obtain the number of keywords numZ1 in Z1 and the number of keywords numZ2 in Z2. When numZ2≤numZ1, at the end of the third unit sampling period, interrupt the interaction between the target user and the artificial intelligence assistant.
[0041] In order to better implement the above method, an artificial intelligence-based computing platform operation and maintenance data analysis system is also proposed;
[0042] The system includes: a database management module, a keyword prediction module, a deviation coefficient calculation module, a feature word overlap calculation module and an interruption management module, wherein the database management module is used to manage the historical data of the answer record, the keyword prediction module is used to predict the keywords included in the content of the user's question again, the deviation coefficient calculation module is used to calculate the deviation coefficient, the feature word overlap calculation module is used to calculate the feature word overlap between the question content, and the interruption management module is used to interrupt the abnormal interaction between the user and the artificial intelligence assistant;
[0043] Furthermore, the database management module includes: a data storage unit, a process management unit and a keyword management unit, wherein the data storage unit is used to store historical data of response records, the process management unit is used to manage task nodes of the processing flow of user requirements, and the keyword management unit is used to manage keywords of task nodes.
[0044] Furthermore, the keyword prediction module includes: an occurrence probability calculation unit, a probability cumulative value calculation unit and a keyword selection unit, wherein the occurrence probability calculation unit is used to calculate the occurrence probability of the keyword, the probability cumulative value calculation unit is used to calculate the probability cumulative value of the keyword, and the keyword selection unit is used to add key words whose probability cumulative value is greater than or equal to the probability management threshold into the keyword prediction set.
[0045] Furthermore, the deviation coefficient calculation module includes: a keyword prediction unit, a keyword similarity calculation unit and a deviation coefficient calculation unit, wherein the keyword prediction unit is used to obtain a keyword prediction set, the keyword similarity calculation unit is used to calculate the first keyword similarity and the second keyword similarity, and the deviation coefficient calculation unit is used to calculate the deviation coefficient between question contents.
[0046] Furthermore, the feature word overlap calculation module includes: a feature word difference set management unit, a new feature word number calculation unit, a feature word set management unit and a feature word overlap calculation unit, wherein the feature word difference set management unit is used to obtain a feature word difference set, the new feature word number calculation unit is used to calculate the number of new feature words, the feature word set management unit is used to manage a set of feature words corresponding to each question content, and the feature word overlap calculation unit is used to calculate the feature word overlap.
[0047] Furthermore, the interruption management module includes: a retention coefficient calculation unit, a keyword difference comparison unit and an interruption execution unit, wherein the retention coefficient calculation unit is used to calculate the retention coefficient, the keyword difference comparison unit is used to compare the differences in task node keywords, and the interruption execution unit is used to execute the interaction between the user and the artificial intelligence assistant to interrupt.
[0048] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: the present invention predicts the keywords in the user's next question in the customer service system through relevant processes and historical data, compares the similarity between the user's next question and the previous question and the predicted result, determines whether the answer provided by the artificial intelligence assistant is helpful to the user, and refers to the relevant feature words in the user's question content to determine whether the user repeatedly inquires about the agreed topic, and further determines the effectiveness of the artificial intelligence assistant's answer. When the effectiveness of the artificial intelligence assistant's answer is low, the communication between the artificial intelligence assistant and the user is interrupted to reduce the generation of invalid communication among users. BRIEF DESCRIPTION OF THE DRAWINGS
[0049] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:
[0050] Figure 1 This is a schematic diagram of the structure of a computing power platform operation and maintenance data analysis system based on artificial intelligence in the present invention patent;
[0051] Figure 2 It is a flow chart of a method for analyzing operation and maintenance data of a computing power platform based on artificial intelligence in the patent of this invention. DETAILED DESCRIPTION
[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0053] See also Figure 1 and Figure 2 , the present invention provides a technical solution:
[0054] Step S100: Obtain the historical records of processing answering tasks from the customer service answering record database, obtain the task nodes in the answering task process, extract the node features in each task node from the historical data, and obtain the keywords corresponding to each task node;
[0055] Wherein, step S100 includes:
[0056] Step S101: obtaining the user's needs from the customer service response record, arranging the interaction data of the same user in chronological order, obtaining an interaction sample sequence, the interaction data including text, voice or video interaction data, and collecting the interaction sample sequences of different users with the same user need to obtain an interaction sample set corresponding to the user need;
[0057] Step S102: Obtaining task nodes in the processing flow of each user demand, extracting interaction data corresponding to each task node from the interaction sample set, and extracting feature words corresponding to the task node from the interaction data, wherein the feature words are keywords corresponding to the task node;
[0058] The interactive sample sequence is divided into several intervals, and the intervals in the same question-and-answer stage are aligned by labeling or classification. Each question-and-answer stage is equivalent to a task node.
[0059] Step S200: In an interactive system that uses an artificial intelligence assistant for answering, collect the historical records of data interaction between the user and the artificial intelligence assistant, extract the pattern of the user asking questions again after the artificial intelligence assistant answers, and predict the keywords included in the content of the user's questions again;
[0060] Wherein, step S200 includes:
[0061] Step S201: Obtain a first task node and a second task node from a processing flow of any task requirement, where the first task node and the second task node are two adjacent task nodes, and obtain a first keyword set composed of characteristic words corresponding to the first task node and a second keyword set composed of characteristic words corresponding to the second task node;
[0062] Step S202: collect all interaction sample sequences including the first task node and the second task node in the historical records to form a target sample sequence set, obtain any keyword prompt1 from the first keyword set, obtain the number K1 of interaction sample sequences including prompt1 in all interaction sample sequences, obtain any keyword prompt2 from the second keyword set, and when an interaction sample sequence includes prompt1 and prompt2, construct a target keyword pair: prompt1→prompt2;
[0063] Step S203: obtaining the number K2 of target keyword pairs in the target sample sequence set, and calculating the occurrence probability of keyword prompt2 α=K2 / K1;
[0064] Step S204: Obtain the target user's interaction data at the first task node, collect the keywords in the interaction data into a set H, and obtain the cumulative occurrence probability of any keyword corresponding to the second task node, where the probability cumulative value of the i-th keyword corresponding to the second task node is R i , Among them, α j represents: the probability of occurrence of any keyword of the i-th node in the second task when the j-th keyword in the set H appears, and b represents the number of keywords in H;
[0065] For example, the keywords (prompt11, prompt12, prompt13) in H are collected once;
[0066] The keywords and occurrence probabilities of prompt11 corresponding to the second task node are: (prompt21, 30%), (prompt23, 15%);
[0067] The keywords and occurrence probabilities of prompt12 corresponding to the second task node are: (prompt21, 5%), (prompt22, 25%), (prompt23, 23%);
[0068] The keywords and occurrence probabilities of prompt13 corresponding to the second task node are: (prompt21, 40%), (prompt22, 20%), (prompt23, 44%);
[0069] The corresponding relationship between H and the keywords and occurrence probabilities of the second task node is (prompt21, 0.75), (prompt22, 0.45), (prompt23, 0.82);
[0070] Step S205: Collect the probability cumulative values of the keywords corresponding to the second task nodes corresponding to the keywords in H, set the probability management threshold, and add the key words with probability cumulative values greater than or equal to the probability management threshold into the keyword prediction set;
[0071] Step S300: a user currently communicating with the artificial intelligence assistant is set as a target user, keywords included in the first predicted content are obtained according to the first question content of the target user, and a deviation coefficient between the first question content and the first predicted content is calculated for the second question content;
[0072] Wherein, step S300 includes:
[0073] Step S301: obtaining any question data of the target user at the first task node as the first question content, collecting the keywords in the first question content into a set H1, and obtaining the keyword prediction set Q1 corresponding to the second task node according to H1;
[0074] Step S302: Obtain a piece of question data after the first question content as the second question data, collect the keywords in the second question content into a set H2, calculate the similarity between the keywords in H2 and the keywords in H1, record it as the first keyword similarity η1, calculate the similarity between the keywords in H2 and the keywords in Q1, record it as the second keyword similarity η2;
[0075] The similarity of keywords can be achieved through relevant keyword similarity algorithms, such as word embedding models, semantic network models or pre-trained semantic models;
[0076] Step S303: Calculate the deviation coefficient y corresponding to the first question content and the second question content, y=η1-η2;
[0077] Step S400: The target user's multiple questions are grouped into a question sequence, and the question sequence is divided by unit sampling period. When the number of newly added feature words in the second unit sampling period is less than that in the first unit sampling period, the feature word overlap between the question contents in the second unit sampling period is calculated;
[0078] Wherein, step S400 includes:
[0079] Step S401: Extract features from any question content and obtain feature words corresponding to each question content. In the question sequence, the feature words corresponding to the pth question content form a set E p , the feature words corresponding to the p+1th question content form a set E p+1 , calculate the feature word difference set D, D = E p+1 -E p , get the number of feature words numD in D;
[0080] Step S402: summing the number of feature words in all feature word difference sets in any unit sampling period, and calculating the average value of the summation results, where the average value is the number of newly added feature words corresponding to any unit period;
[0081] Step S403: Arrange the feature words corresponding to each question content in any unit sampling period to obtain L, L: f1, f2, f3, ... fn, where f1, f2, f3, ... and fn respectively represent the set of feature words corresponding to the 1st, 2nd, 3rd, ... and nth question content in any unit sampling period;
[0082] Step S404: Take the intersection of all sets in L to obtain set L1, take the union of all sets in L to obtain set L2, obtain the number of feature words numl1 in set L1, the number of feature words numl2 in set L2, and calculate the feature word overlap degree β in any unit sampling period, β=numL1 / numL2;
[0083] Step S500: Obtain the deviation coefficient and feature word overlap of the question content in a unit sampling period, calculate the retention coefficient, and when the retention coefficient corresponding to the second unit sampling period is greater than the management threshold, obtain the current task node of the target user, compare the question content in the third unit sampling period with the second unit sampling period, and when the question content includes no increase in the keywords of the next task node, at the end of the third unit sampling period, interrupt the interaction between the target user and the artificial intelligence assistant;
[0084] Wherein, step S500 includes:
[0085] Step S501: sum the deviation coefficients in the second unit sampling period to obtain the first deviation value Y, calculate the retention coefficient reten of the second unit sampling period, reten=Y×β, obtain the management threshold μ, and when reten≥μ, obtain the keywords corresponding to the second task node in all question contents in the first unit sampling period to form a set G1, and obtain the keywords corresponding to the second task node in all question contents in the second unit sampling period to form a set G2;
[0086] Step S502: Calculate the first keyword difference set Z1, Z1 = G2-G1, obtain the keywords in all the question contents in the third unit sampling period, form a set G3, and calculate the second keyword difference set Z2, Z2 = G3-G2;
[0087] Step S503: Obtain the number of keywords numZ1 in Z1 and the number of keywords numZ2 in Z2. When numZ2≤numZ1, at the end of the third unit sampling period, interrupt the interaction between the target user and the artificial intelligence assistant.
[0088] The system includes:
[0089] Database management module, keyword prediction module, deviation coefficient calculation module, feature word coincidence calculation module and interruption management module;
[0090] The database management module is used to manage the historical data of the answer record, wherein the database management module includes: a data storage unit, a process management unit and a keyword management unit, wherein the data storage unit is used to store the historical data of the answer record, the process management unit is used to manage the task nodes of the processing flow of the user's needs, and the keyword management unit is used to manage the keywords of the task nodes;
[0091] The keyword prediction module is used to predict the keywords included in the content of the user's question again, wherein the keyword prediction module includes: an appearance probability calculation unit, a probability cumulative value calculation unit and a keyword selection unit, wherein the appearance probability calculation unit is used to calculate the appearance probability of the keyword, the probability cumulative value calculation unit is used to calculate the probability cumulative value of the keyword, and the keyword selection unit is used to add the key words whose probability cumulative value is greater than or equal to the probability management threshold into the keyword prediction set;
[0092] The deviation coefficient calculation module is used to calculate the deviation coefficient, wherein the deviation coefficient calculation module includes: a keyword prediction unit, a keyword similarity calculation unit and a deviation coefficient calculation unit, wherein the keyword prediction unit is used to obtain a keyword prediction set, the keyword similarity calculation unit is used to calculate the first keyword similarity and the second keyword similarity, and the deviation coefficient calculation unit is used to calculate the deviation coefficient between the question contents;
[0093] The feature word overlap calculation module is used to calculate the feature word overlap between question contents, wherein the feature word overlap calculation module includes: a feature word difference set management unit, a newly added feature word quantity calculation unit, a feature word set management unit and a feature word overlap calculation unit, wherein the feature word difference set management unit is used to obtain a feature word difference set, the newly added feature word quantity calculation unit is used to calculate the number of newly added feature words, the feature word set management unit is used to manage a set of feature words corresponding to each question content, and the feature word overlap calculation unit is used to calculate the feature word overlap;
[0094] Among them, the interruption management module is used to interrupt the abnormal interaction between the user and the artificial intelligence assistant, wherein the interruption management module includes: a retention coefficient calculation unit, a keyword difference comparison unit and an interruption execution unit, wherein the retention coefficient calculation unit is used to calculate the retention coefficient, the keyword difference comparison unit is used to compare the differences in the keywords of the task nodes, and the interruption execution unit is used to interrupt the interaction between the user and the artificial intelligence assistant.
[0095] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.
[0096] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A computing platform operation and maintenance data analysis method based on artificial intelligence, characterized in that: The method comprises the following steps: Step S100: Obtain the historical records of processing answering tasks from the customer service answering record database, obtain the task nodes in the answering task process, extract the node features in each task node from the historical data, and obtain the keywords corresponding to each task node; Step S200: In an interactive system that uses an artificial intelligence assistant for answering, collect the historical records of data interaction between the user and the artificial intelligence assistant, extract the pattern of the user asking questions again after the artificial intelligence assistant answers, and predict the keywords included in the content of the user's questions again; Step S300: a user currently communicating with the artificial intelligence assistant is set as a target user, keywords included in the first predicted content are obtained according to the first question content of the target user, and a deviation coefficient between the first question content and the first predicted content is calculated for the second question content; Step S300 includes: Step S301: obtaining any question data of the target user at the first task node as the first question content, collecting the keywords in the first question content into a set H1, and obtaining the keyword prediction set Q1 corresponding to the second task node according to H1; Step S302: obtaining a piece of question data after the first question content as the second question data, collecting the keywords in the second question content into a set H2, calculating the similarity between the keywords in H2 and the keywords in H1, recorded as the first keyword similarity η1, calculating the similarity between the keywords in H2 and the keywords in Q1, recorded as the second keyword similarity η2; Step S303: Calculate the deviation coefficient y corresponding to the first question content and the second question content, y=η1-η2; Step S400: The target user's multiple questions are grouped into a question sequence, and the question sequence is divided by unit sampling period. When the number of newly added feature words in the second unit sampling period is less than that in the first unit sampling period, the feature word overlap between the question contents in the second unit sampling period is calculated; Step S400 includes: Step S401: Extract features from any question content and obtain feature words corresponding to each question content. In the question sequence, the feature words corresponding to the pth question content form a set E p , the feature words corresponding to the p+1th question content form a set E p+1 , calculate the feature word difference set D, D = E p+1 -E p , get the number of feature words numD in D; Step S402: summing the number of feature words in all feature word difference sets in any unit sampling period, and calculating an average value of the summation results, where the average value is the number of newly added feature words corresponding to any unit period; Step S403: Arrange the feature words corresponding to each question content in any unit sampling period to obtain L, L: f1, f2, f3, ... fn, where f1, f2, f3, ... and fn respectively represent the set of feature words corresponding to the 1st, 2nd, 3rd, ... and nth question content in any unit sampling period; Step S404: Take the intersection of all sets in L to obtain set L1, take the union of all sets in L to obtain set L2, obtain the number of feature words numl1 in set L1, the number of feature words numl2 in set L2, and calculate the feature word overlap degree β in any unit sampling period, β=numL1 / numL2; Step S500: Obtain the deviation coefficient and feature word overlap of the question content in a unit sampling period, calculate the retention coefficient, and when the retention coefficient corresponding to the second unit sampling period is greater than the management threshold, obtain the current task node of the target user, compare the question content in the third unit sampling period with the second unit sampling period, and when the question content includes no increase in the keywords of the next task node, at the end of the third unit sampling period, interrupt the interaction between the target user and the artificial intelligence assistant; Step S500 includes: Step S501: sum the deviation coefficients in the second unit sampling period to obtain the first deviation value Y, calculate the retention coefficient reten of the second unit sampling period, reten=Y×β, obtain the management threshold μ, and when reten≥μ, obtain the keywords corresponding to the second task node in all question contents in the first unit sampling period to form a set G1, and obtain the keywords corresponding to the second task node in all question contents in the second unit sampling period to form a set G2; Step S502: Calculate the first keyword difference set Z1, Z1 = G2-G1, obtain the keywords in all the question contents in the third unit sampling period, form a set G3, and calculate the second keyword difference set Z2, Z2 = G3-G2; Step S503: Obtain the number of keywords numZ1 in Z1 and the number of keywords numZ2 in Z2. When numZ2≤numZ1, at the end of the third unit sampling period, interrupt the interaction between the target user and the artificial intelligence assistant.
2. The method for analyzing computing power platform operation and maintenance data based on artificial intelligence according to claim 1, characterized in that: Step S100 includes: Step S101: obtaining a user's user demand from a customer service response record, arranging the interaction data of the same user in chronological order to obtain an interaction sample sequence, wherein the interaction data includes text, voice or video interaction data, and collecting the interaction sample sequences of different users with the same user demand to obtain an interaction sample set corresponding to the user demand; Step S102: Obtain the task nodes in the processing flow of each user demand, extract the interaction data corresponding to each task node from the interaction sample set, and extract the characteristic words corresponding to the task node from the interaction data, wherein the characteristic words are the keywords corresponding to the task node.
3. The method for analyzing computing power platform operation and maintenance data based on artificial intelligence according to claim 2, characterized in that: Step S200 includes: Step S201: Obtain a first task node and a second task node from a processing flow of any task requirement, wherein the first task node and the second task node are two adjacent task nodes, and obtain a first keyword set composed of characteristic words corresponding to the first task node and a second keyword set composed of characteristic words corresponding to the second task node; Step S202: collect all interaction sample sequences including the first task node and the second task node in the historical records to form a target sample sequence set, obtain any keyword prompt1 from the first keyword set, obtain the number K1 of interaction sample sequences including prompt1 in all interaction sample sequences, obtain any keyword prompt2 from the second keyword set, and when an interaction sample sequence includes prompt1 and prompt2, construct a target keyword pair: prompt1→prompt2; Step S203: obtaining the number K2 of target keyword pairs in the target sample sequence set, and calculating the occurrence probability of keyword prompt2 α=K2 / K1; Step S204: Obtain the target user's interaction data at the first task node, collect the keywords in the interaction data into a set H, and obtain the cumulative occurrence probability of any keyword corresponding to the second task node, where the probability cumulative value of the i-th keyword corresponding to the second task node is R i , Among them, α j represents: the probability of occurrence of any keyword of the i-th node in the second task when the j-th keyword in the set H appears, and b represents the number of keywords in H; Step S205: Gather the probability cumulative values of the keywords corresponding to the second task nodes corresponding to the keywords in H, set a probability management threshold, and add key words with probability cumulative values greater than or equal to the probability management threshold into the keyword prediction set.
4. A computing power platform operation and maintenance data analysis system applied to the computing power platform operation and maintenance data analysis method based on artificial intelligence as described in any one of claims 1 to 3, characterized in that: The system includes the following modules: a database management module, a keyword prediction module, a deviation coefficient calculation module, a feature word coincidence calculation module and an interruption management module, wherein the database management module is used to manage the historical data of the answer record, the keyword prediction module is used to predict the keywords included in the content of the user's next question, the deviation coefficient calculation module is used to calculate the deviation coefficient, the feature word coincidence calculation module is used to calculate the feature word coincidence between the question contents, and the interruption management module is used to interrupt the abnormal interaction between the user and the artificial intelligence assistant.
5. The computing power platform operation and maintenance data analysis system according to claim 4, characterized in that: The database management module includes: a data storage unit, a process management unit and a keyword management unit, wherein the data storage unit is used to store historical data of the response record, the process management unit is used to manage the task nodes of the processing flow of user requirements, and the keyword management unit is used to manage the keywords of the task nodes; The keyword prediction module includes: an occurrence probability calculation unit, a probability cumulative value calculation unit and a keyword selection unit, wherein the occurrence probability calculation unit is used to calculate the occurrence probability of the keyword, the probability cumulative value calculation unit is used to calculate the probability cumulative value of the keyword, and the keyword selection unit is used to add key words whose probability cumulative value is greater than or equal to the probability management threshold into the keyword prediction set.
6. The computing power platform operation and maintenance data analysis system according to claim 4, characterized in that: The deviation coefficient calculation module includes: a keyword prediction unit, a keyword similarity calculation unit and a deviation coefficient calculation unit, wherein the keyword prediction unit is used to obtain a keyword prediction set, the keyword similarity calculation unit is used to calculate the first keyword similarity and the second keyword similarity, and the deviation coefficient calculation unit is used to calculate the deviation coefficient between question contents; The feature word overlap calculation module includes: a feature word difference set management unit, a new feature word number calculation unit, a feature word set management unit and a feature word overlap calculation unit, wherein the feature word difference set management unit is used to obtain a feature word difference set, the new feature word number calculation unit is used to calculate the number of new feature words, the feature word set management unit is used to manage a set of feature words corresponding to each question content, and the feature word overlap calculation unit is used to calculate the feature word overlap.
7. The computing power platform operation and maintenance data analysis system according to claim 4, characterized in that: The interruption management module includes: a retention coefficient calculation unit, a keyword difference comparison unit and an interruption execution unit, wherein the retention coefficient calculation unit is used to calculate the retention coefficient, the keyword difference comparison unit is used to compare the differences in task node keywords, and the interruption execution unit is used to execute the interaction between the user and the artificial intelligence assistant to interrupt.
Citation Information
Patent Citations
Intelligent customer service response method and system
CN110162611A
Intelligent question and answer interaction system and method based on visual flow chart
CN113297361A
